## Abstract

Since the mid-1990s, the United States has witnessed a dramatic rise in drug overdose mortality. Educational gradients in life expectancy widened over the same period, and drug overdose likely plays a role in this widening, particularly for non-Hispanic whites. The contemporary drug epidemic is distinctive in terms of its scope, the nature of the substances involved, and its geographic patterning, which influence how it impacts different education groups. I use vital statistics and National Health Interview Survey data to examine the contribution of drug overdose to educational gradients in life expectancy from 1992–2011. I find that over this period, years of life lost due to drug overdose increased for all education groups and for both males and females. The contribution of drug overdose to educational gradients in life expectancy has increased over time and is greater for non-Hispanic whites than for the population as a whole. Drug overdose accounts for a sizable proportion of the increases in educational gradients in life expectancy, particularly at the prime adult ages (ages 30–60), where it accounts for 25 % to 100 % of the widening in educational gradients between 1992 and 2011. Drug overdose mortality has increased more rapidly for females than for males, leading to a gender convergence. These findings shed light on the processes driving recent changes in educational gradients in life expectancy and suggest that effective measures to address the drug overdose epidemic should take into account its differential burden across education groups.

## Introduction

Since the mid-1990s, the United States has witnessed a dramatic rise in drug overdose mortality. As of 2016, deaths from drug overdose exceed deaths from motor vehicle accidents and homicide (Xu et al. 2016). Drug overdose has also been shown to be an important contributor to Americans’ life expectancy shortfall relative to other high-income countries (Ho 2013). Between 1994 and 2014, the age-standardized death rate from drug overdose more than tripled, from 4.8 to 14.7 deaths per 100,000 (CDC/NCHS 2015). Over the same period, death rates from motor vehicle accidents and homicides declined by 30 % to 45 %. The drug overdose epidemic has been primarily driven by prescription opioid painkillers, although deaths in which illicit opioids such as heroin and fentanyl are implicated are on the rise, culminating in the “twin” epidemics of opioids and heroin (Paulozzi et al. 2011; Rudd et al. 2016).

Several studies have documented widening educational gradients in mortality over time (Elo and Preston 1996; Hendi 2015; Meara et al. 2008; Miech et al. 2011). Studies focusing on trends in the 1990s and 2000s highlighted stagnating life expectancy gains among less-educated subgroups, particularly for non-Hispanic white women, and continued improvements among more-educated subgroups as drivers of widening educational gradients (Hendi 2015; Meara et al. 2008).

Stagnating life expectancy gains among the less-educated coincided with the takeoff of drug overdose mortality in the mid- to late 1990s following FDA approval of the opioid pain reliever OxyContin in December 1995. Increases in drug overdose mortality were especially steep in the early 2000s. Although drug overdose death rates rose for most groups, increases were largest for non-Hispanic whites. Between 1999 and 2014, age-standardized death rates from drug overdose increased by 211.5 %, 40.0 %, and 21.8 % for non-Hispanic whites, non-Hispanic blacks, and Hispanics, respectively (CDC/NCHS 2015). Non-Hispanic whites now have higher drug overdose death rates than non-Hispanic blacks and Hispanics. The widening of the educational gradients in life expectancy has also been particularly pronounced for non-Hispanic whites. Thus, it seems likely that the drug overdose epidemic may play a role in the widening of these gradients.

However, no prior studies have examined the contribution of drug overdose to life expectancy differences by education or to the widening of these differences over time. This study uses data from the National Health Interview Survey (NHIS) linked to the National Death Index (NDI), combined with data from the National Vital Statistics System (NVSS), to examine the contribution of drug overdose to educational gradients in life expectancy. I use cause-deleted life table methods to estimate (1) the years of life lost due to drug overdose by sex and education in four periods: 1992–1996, 1997–2001, 2002–2006, and 2007–2011; (2) the contribution of drug overdose mortality to educational gradients in life expectancy in each of these periods; and (3) the contribution of drug overdose mortality to changes over time in these gradients.

## Background

The contemporary drug overdose epidemic is distinctive in terms of its scope, its geographic patterning, and the nature of the substances involved. Deaths by poisoning, a large fraction of which are drug overdose deaths, reached 51,966 in 2014, and are now the leading cause of injury-related deaths in the United States, accounting for more than one-quarter (26.0 %) of injury deaths (Kochanek et al. 2016). The respective figures for 1997 were 17,692 deaths and 12.1 % (Hoyert et al. 1999). Before the current epidemic, a positive association existed between the level of urbanization and drug overdose mortality. Drug overdose deaths were concentrated in large cities, such as Baltimore and New York City, which were connected to international drug smuggling networks based in Europe and Turkey. Large central metros had the highest drug overdose death rates—more than 2.5 times the death rates of the most rural areas, which had the lowest death rates (Paulozzi and Xi 2008). In recent decades, however, this association disappeared. Rural areas and suburbs, which were not profitable markets for illegal drugs prior to the current epidemic, have witnessed the largest increases in drug overdose mortality, and people in rural counties are now more likely to overdose than people in big cities. This change is partly related to a fundamental shift in the nature of the substances involved.

In contrast to the heroin and cocaine epidemics of the 1970s and 1980s, which involved illicit drugs, the current epidemic is driven primarily by legally prescribed drugs. Since the early 2000s, opioid painkillers—particularly oxycodone—have become the leading cause of drug overdose deaths (Paulozzi et al. 2011). In 2007, they were implicated in more overdose deaths than heroin and cocaine combined (Okie 2010). Although the use of heroin and other illicit drugs is also increasing and contributing to drug overdose mortality, this has been fueled in large part by users switching from prescription opioids.1 In 2002–2011, nearly 80 % of new heroin users aged 12–49 reported abusing prescription opioids prior to using heroin (Muhuri et al. 2013). In contrast, in the three decades prior to the 2000s, the great majority of heroin users reported that heroin was their first opioid of abuse (Cicero et al. 2014).

The roots of the prescription drug epidemic lie in the pain management revolution. Before the 1980s, opioid painkillers were rarely prescribed except for terminally ill cancer patients with chronic pain, and their use was strictly regulated because of fears of addiction. Over the next two decades, however, a sea change in pain assessment and treatment occurred. Pain went from being a neglected phenomenon to being hailed as the “fifth vital sign” starting in 1996, and pain assessments were incorporated in routine health measurements alongside pulse, blood pressure, and temperature (American Pain Society 1999). Several factors contributed to this change, including (1) the patients’ rights movements, which espoused that freedom from pain is a universal human right and that pain was drastically undertreated in the United States; (2) doctors’ faith in medical progress that we had reached a point where pain could be safely treated by opioid painkillers; (3) campaigns to destigmatize those painkillers; and (4) Purdue Pharma’s aggressive marketing of the opioid painkiller OxyContin as low risk and nonaddictive.

This pain revolution culminated in an astronomical rise in the prescribing of opioid painkillers, reflected in sharp increases in drug-related emergency department visits, substance abuse treatment admissions, and drug overdose deaths. Sales of opioid painkillers quadrupled between 1999 and 2010 (Paulozzi et al. 2011). This increase in prescribing was not, however, accompanied by increases in the prevalence of patients reporting pain. Between 2000 and 2010, the proportions of emergency department visits and outpatient office visits for which pain was the primary symptom or diagnosis remained stable (Chang et al. 2014; Daubresse et al. 2013). However, prescriptions of opioid painkillers increased dramatically while the role of other (non-opioid) therapies was reduced. Factors contributing to the unprecedented amount and diversion of painkillers entering the population include “pill mills,” clinics where doctors prescribe extremely large quantities of painkillers for cash; “sponsors” who drive groups of addicts to pain clinics in return for most of the pills they receive, which they in turn resell; and doctor shopping, in which patients obtain prescriptions from multiple doctors (McDonald and Carlson 2013; Rigg et al. 2010; Temple 2015). In 2007, the United States alone accounted for 83 % and 99 % of global consumption of the opium derivatives oxycodone (the active ingredient in OxyContin) and hydrocodone, respectively (International Narcotics Control Board 2009). Other high-income countries have not experienced such severe prescription painkiller epidemics, which may be related to their more conservative attitudes toward painkiller prescribing and less aggressive treatment of chronic disease. Compared with the more fragmented U.S. health care system, health care systems in other countries have greater coordination of care and fewer opportunities for doctor shopping, for-profit clinics, and pill mills.

### Drug Overdose and Educational Differences

The linkages between socioeconomic status (SES)—of which education is one indicator—and drug use vary across time and space. The typical pattern is for elites2 to begin experimenting with drugs, which has occurred in the past with substances such as cigarettes, cocaine, and opium. After such use diffuses and becomes more widespread in the population, it loses its social cachet, more attention is paid to its negative consequences, and public disapproval mounts (Collins 2005; Ho and Fenelon 2015; Pampel 2001). At this point, use declines and may become stigmatized, criminalized, and/or concentrated among low-status groups. In time, these events are forgotten, and the cycle repeats (Klaue 1999; Musto 1991). These patterns are consistent with hypotheses from the diffusion of innovations literature, which suggest that the more-educated are early adopters and that between-group differences in health behaviors are a function of the stage of diffusion (Pampel 2001; Pampel et al. 2010). For example, Pampel (2001) found that the highly educated, among whom smoking was the first to diffuse, had smaller sex differences in cigarette smoking. Thus, one hypothesis, based on the prior literature and historical observation of the linkages between status and patterns of drug use, is that the more-educated may have higher drug overdose death rates in the initial stages of the epidemic.

Several distinctive features of the contemporary drug overdose epidemic may influence how it impacts different education groups: the role of the health care system, its rise in rural areas, and its propagation through social networks. First, use of opioid painkillers requires at some point an interaction with the health care system because opioids are legally prescribed, controlled substances (even if they are eventually diverted). The more-educated are more likely to have greater access to the health care system and better insurance policies covering the cost of prescription drugs. On the other hand, cost may be less of a barrier for low-educated individuals who qualify for Medicaid or programs such as Supplemental Security Income, which usually include Medicaid benefits. The low-educated are also more likely to be targeted by and more vulnerable to sponsors who take them to pain clinics, which are easier to access than other parts of the health care system and which dispense large amounts of painkillers for cash.

Following the fundamental cause perspective, drug overdose can be conceptualized as a proximate risk factor or cause of death. This perspective suggests that to identify the root sources of health disparities, scholars should look to the broader social conditions that form the context in which proximate behaviors such as drug use, diversion, and abuse take place (Link and Phelan 1995). Long-term structural changes in the labor market, such as deindustrialization, have resulted in reduced job opportunities and economic prospects for the less-educated. Relative to the more-educated, the low-educated are more likely to be in poverty, which is a risk factor for drug abuse, and the aforementioned structural changes may incentivize them to participate in the informal economies that emerged around opioid painkillers, particularly in rural areas (Keyes et al. 2014). One study found that OxyContin use was associated with greater social capital in rural Appalachia, hypothesizing that OxyContin may be a means of gaining greater social stature and economic advantages (Jonas et al. 2012).

The proliferation of prescription opioids in rural areas has other implications for educational differences. In general, education levels are lower in rural areas. These areas experienced higher unemployment, higher levels of poverty, and slower recovery after the Great Recession, which may increase vulnerability to drug use. Individuals in rural areas are less likely to live within close proximity to drug treatment programs, which are overburdened and in short supply. Travel distances for emergency responders and to emergency departments are also greater in rural than in more urban areas. This distance is critical for death from drug overdose, which often occurs as a result of respiratory and/or central nervous system depression. Deaths can often be prevented if patients receive life support and naloxone, which reverses the effects of opioids if administered soon after an overdose. Reaching patients and providing treatment immediately following an overdose is more challenging in rural areas because of lower population density, longer response times, personnel shortages, lack of advanced training opportunities for emergency medical services (EMS) providers, and antiquated equipment (Faul et al. 2015). A recent study suggested that naloxone is underused in rural areas, and this may be related to the greater proportion of EMS personnel in rural areas that are certified as basic versus intermediate or paramedic, given that only 12 states allow EMT-basic personnel to administer naloxone (Faul et al. 2015). Less-educated individuals are more likely to be exposed to these adverse conditions because of their greater likelihood of living in rural areas. More-educated individuals are more likely to live in areas with higher quality, more rapid, and more comprehensive provision of emergency medical services, and they may also be more knowledgeable about the importance of seeking immediate treatment following an overdose.

The fundamental cause perspective focuses on the enduring linkages between SES and disease, highlighting the critical role of resources in maintaining these associations (Link and Phelan 1995). In the case of drug overdose, the better-educated may possess more resources to combat drug addiction and overdose, including greater access to financial resources, drug treatment programs, and social support networks that can be mobilized to counter drug use. Nonmedical use of prescription opioids is heavily network-based: Jones et al. (2014) documented that for 70.6 % of those reporting nonmedical use of opioid painkillers in the past year, friends and relatives were the source of the drugs. The less-educated may be more likely to be connected to individuals who have been prescribed painkillers or who participate in drug diversion through their social networks. For example, in rural communities, the nature and importance of family and community networks tend to result in wide social networks consisting of strong ties, facilitating the diversion and spread of prescription opioids (Keyes et al. 2014). Less-educated individuals are more likely to be in poorer health; work in blue-collar occupations, such as mining and manufacturing, where they may incur work-related injuries and disabilities; and have more chronic conditions, particularly pain-related conditions. This increases the likelihood of their being prescribed opioid painkillers and high dosages of opioids, which are associated with greater risks of fatal drug overdose (Bohnert et al. 2011).

Thus, the burden of drug overdose mortality is likely to be differentially distributed across education groups. However, few studies have examined this question. This article seeks to estimate the magnitude of educational differences in drug overdose mortality and demonstrate how they have changed over time. These analyses shed light on the processes affecting the size of and trends in contemporary educational gradients in life expectancy in the United States.

## Data

I use vital statistics and survey data to estimate the burden of drug overdose mortality by education. First, I estimate the fraction of all deaths due to drug overdose (DO) using the CDC/NCHS Multiple Cause-of-Death (MCD) micro-data files:
$risepDO=DisepDODisepTotal,$
1
where D is the number of deaths, i is age (25–29, . . . , 80–84, 85+), s is sex, e is education (less than high school, high school completion, some college, college or more), and p is period. The “some college” and “college or more” categories correspond to having one to three years or four or more years of college, respectively, defined based on years of education (for states using the 1989 revision of the death certificate, NHIS prior to 1996) or highest degree attained (for states using the 2003 revision of the death certificate, NHIS from 1997 onward).

Drug overdose deaths are defined according to the standard NCHS category of drug poisoning deaths (Rudd et al. 2016; Warner et al. 2011).3 This is the most precise and reliable definition of drug overdose deaths because (1) it includes deaths from both legal and illegal drugs; (2) it excludes drug-related deaths unlikely to be related to opioids (e.g., drug-induced obesity or poisonings due to pesticides), whose inclusion would lead to overestimates of drug overdose mortality; and (3) it includes drug-related deaths regardless of intentionality, including accidents, suicides, and deaths of undetermined intent (including only a subset of these deaths such as accidental poisoning would lead to underestimates of drug overdose mortality).

Completeness of education reporting on death certificates has varied over time and across states (Rostron et al. 2010). I conduct a sensitivity analysis by calculating life expectancy in the absence of drug overdose using data from all states with any education reporting, from states with at least 90 % completeness of education reporting, and from the 34 states with at least 90 % completeness since 1992. Differences across these variants are minimal and never exceed 0.01 years (Table 4 in the  appendix). The results in the main text are based on all available data.

Second, I estimate all-cause death rates by age, sex, education, and period (misep) from weighted counts of deaths and quarter-years of exposure using the NHIS (Minnesota Population Center and State Health Access Data Assistance Center 2015). I use the NHIS because calculating education-specific death rates from death certificates and census population counts is problematic owing to dual data-source bias from education misreporting on death certificates (Hendi forthcoming; Rostron et al. 2010; Sorlie and Johnson 1996). The NHIS is representative of the U.S. civilian noninstitutionalized population. I use data on individuals surveyed between 1986 and 2009 who contributed deaths and person-years during four periods: one prior to the takeoff in drug overdose mortality (1992–1996) and three subsequent periods (1997–2001, 2002–2006, and 2007–2011). Individuals are allowed to contribute a maximum of 10 years of exposure, or are censored at time of death or on December 31, 2011.

Finally, I apply the ratios of drug overdose to total deaths from vital statistics to the death rates from the NHIS to obtain drug overdose death rates by age, sex, education, and period:
$misepDO=risepDO×misep.$
2

Drug overdose death rates by education cannot be calculated directly from the NHIS for several reasons. First, the cause-of-death categories available in the NHIS do not allow for precise identification of drug overdose deaths because these categories include causes that one would like to exclude (e.g., motor vehicle accidents) and exclude causes that one would like to include (e.g., opioid poisoning deaths of undetermined intent). Second, even if one were to use these broad categories, the number of deaths would be too small to derive stable estimates.4 Thus, the combination of vital statistics and survey data used in this article is the most reliable way to calculate nationally representative, education-specific drug overdose death rates.

## Methods

I use standard life table methods to calculate life expectancies by sex, education, and period (esep) using all-cause death rates from the NHIS (misep) and nax values of 2.5. The death rates are converted to probabilities of dying (pisep) to construct these life tables.

Next, I calculate life expectancies by sex, education, and period in the absence of drug overdose deaths $esep−DO$ using cause-deleted life tables based on Chiang’s approach (Chiang 1968; Preston et al. 2001). To construct these cause-deleted life tables, nax values are derived using graduation techniques, and the probabilities of dying in the absence of drug overdose are calculated as follows:
$pisep−DO=pisep1−risepDO.$
3

This equation is derived from Chiang’s assumption, which posits that within each age group, the instantaneous death rate in the absence of a particular cause of death is proportional to the observed instantaneous death rate. In this case, the cause of death is drug overdose, and the constant of proportionality is $1−risepDO$. In other words, within each age group, the death rates in the absence of drug overdose have the same shape as the observed death rates but differ in terms of level. Interested readers seeking further details are encouraged to consult Beltrán-Sánchez et al. (2008) and Preston et al. (2001).

I compare two quantities from these life tables in the presence (observed) and absence (counterfactual) of drug overdose: life expectancy at age 25 (e25 vs. $e25−DO$) and years of life lived between ages 30 and 60 (30L30 vs. $L3030−DO$).5 If drug overdose no longer operated as a cause of death, people would live longer than we actually observe ($e25−DO≥e25$ and $L3030−DO≥L3030$). Thus, years of life lost (YLL) due to drug overdose are calculated as follows:
$YLL25+=e25−DO−e25$
4a
$YLL30−60=L3030−DO−L3030$
4b
for each sex-education-period combination (with sep subscripts omitted for clarity). These quantities are used to determine the contributions of drug overdose to educational gradients in life expectancy in each period and to changes in educational gradients in life expectancy over time. For example, the percentage of the educational gradient in life expectancy between high school and college graduates due to drug overdose in a given period is
$e25COL−e25HS−e−DO25COL−e−DO25HSe25COL−e25HS×100.$
5

I calculate these quantities for the total population and for non-Hispanic whites alone because drug overdose death rates are higher6 and educational gradients in life expectancy tend to be larger for non-Hispanic whites.

Cause-deleted life tables compute life expectancy in the absence of all drug overdose mortality. To shed more light on the dynamics linking drug overdose and educational gradients, I examine two additional counterfactuals that answer the following questions:

• (1) What would educational gradients in life expectancy look like if all education groups maintained their drug overdose death rates from the earliest period (1992–1996)?

• (2) What would educational gradients in life expectancy look like if the other three education groups experienced the drug overdose death rates of the college-educated in each period?

In these counterfactuals, mortality rates from all other causes excluding drug overdose are unchanged from the observed rates in each period (i.e., all variation is derived from changes in drug overdose death rates).

## Results

### Drug Overdose Mortality by Education

Drug overdose has become a major cause of death in certain age groups. In 1992, drug overdose accounted for 4.4 % to 5.4 % of all deaths among high school graduates and 2.4 % to 4.5 % of all deaths among college graduates aged 25–34 (see Table 5 in the  appendix). By 2011, these figures reached roughly 20 % for high school graduates and 10 % to 16 % for college graduates. These fractions are higher for non-Hispanic whites than for the total population. The highest fraction was observed for non-Hispanic white females aged 25–29 with less than a high school education in 2010, among whom fully one-third of all deaths were due to drug overdose. In 1992, this figure was only 5.6 %. Overall, however, death rates at these ages are much lower than at other ages, so we examine age-standardized death rates and life expectancy measures.

Figure 1 shows age-standardized drug overdose death rates by education for males (panel a) and females (panel b). In the earliest period, 1992–1996, a negative educational gradient in drug overdose mortality existed among males. Death rates ranged from 3.2 per 100,000 among college graduates to 14.0 per 100,000 among those with less than a high school education. Among females, those with less than a high school education had higher drug overdose death rates, but death rates among the other three education groups were fairly similar. In this period, drug overdose death rates were very similar for the total population and for non-Hispanic whites across all education levels.

The next period, 1997–2001, covers the years directly following the introduction of OxyContin. Compared with the first period, drug overdose death rates are higher for all education groups, with larger increases observed among those with a high school diploma or less and for non-Hispanic whites compared with the total population. A more graded association between education and drug overdose begins to emerge among females.

In the third period, 2002–2006, we continue to see a pattern of steeper increases in drug overdose mortality at the lower end of the education distribution. This is particularly true for non-Hispanic whites, among whom the less than high school group pulls away from the other education groups in this period.

By the most recent period, 2007–2011, drug overdose death rates reached 25.1–29.5 per 100,000 among those with less than a high school education in the total population and 44.2–50.3 per 100,000 among non-Hispanic whites with less than a high school education. These range from two- to fivefold increases compared with the earliest period. Among college graduates, drug overdose death rates increased over time; but in the most recent period, they were still lower than rates observed in the first period for the least-educated.

Drug overdose death rates for males and females converged over this period. Initially, for all three education groups with less than a college degree, drug overdose death rates were roughly two times higher for males than females. By the most recent period, however, the ratio of male to female death rates was much closer to parity. This occurred for both the total population and for non-Hispanic whites. Trends differed among college graduates, whose drug overdose death rates were near parity to begin with. Among college graduates, male drug overdose death rates increased more quickly than female drug overdose death rates, resulting in a peak in the ratio of male to female drug overdose death rates at 1.7 in 2002–2006. Between 2002–2006 and 2007–2011, however, drug overdose death rates increased more rapidly among college-educated females, bringing the ratio back in the range of 1.0–1.2.

### Years of Life Lost Due to Drug Overdose, by Education

Table 1 shows the years of life lost due to drug overdose (YLL) at ages 25+ and between ages 30 and 60 by education for males (panel A) and females (panel B).

In general, there is an inverse association between YLL and education, with those with less than a high school education losing the most years of life from drug overdose, and the college-educated losing the fewest years of life from drug overdose. Differences among those with a high school diploma, some college, and college or more are muted among females until after the takeoff in drug overdose mortality (i.e., the educational gradient in YLL widens starting in the 1997–2001 period for females). In the earliest period, YLL above age 25 by education range from 0.11–0.24 years for males and 0.10–0.18 years for females. In the most recent period, the corresponding ranges are 0.17–0.79 and 0.16–0.75 for males and females, respectively.

Initially, YLL were very similar for the total population and for non-Hispanic whites. Over time, however, they increased more rapidly among non-Hispanic whites than for the total population among the three lower education groups. For example, YLL increased from 0.24 to 0.46 years for males with less than a high school education in the total population between 1992 and 2011. For their non-Hispanic white counterparts, YLL increased from 0.24 to 0.79 years. Among the college-educated, however, YLL remained similar for the total population and for non-Hispanic whites.

The patterns and trends in YLL between ages 30 and 60 are similar to those at ages 25 and older. A large proportion of YLL due to drug overdose mortality above age 25 occurs at ages 30–60. Across all periods and education groups, YLL between ages 30 and 60 accounts for 19 % to 52 % and 17 % to 41 % of all YLL above age 25 for males and females, respectively.

### Contribution of Drug Overdose to Educational Gradients in Life Expectancy

Next, I translate these differences in YLL due to drug overdose into their contributions to educational gradients in life expectancy. Table 2 shows the gradient, or the observed difference in life expectancy at age 25 (or years of life lived between ages 30 and 60) between each education group and the college-educated, and what percentage of the gradient is due to drug overdose in each period. These percentages reflect differences between the observed and cause-deleted life tables; in other words, in the absence of drug overdose, educational gradients in life expectancy would be smaller than we observe.

In the earliest period, drug overdose made minimal contributions (ranging from 0.5 % to 1.7 %) to educational gradients in life expectancy at age 25, and these were similar for the total population and non-Hispanic whites. By the most recent period, the contributions increased to 2.7 % to 6.3 % and were greater among non-Hispanic whites. Interestingly, for life expectancy at age 25, drug overdose often makes a larger relative (percentage) contribution to the differences between the most and intermediate educated (e.g., college-educated vs. some college or high school education) than to the differences between the most- and least-educated (e.g., college-educated vs. less than a high school education). However, the absolute contributions of drug overdose are higher in the latter case because life expectancy differences between the college-educated and those with less than a high school education are larger.

Drug overdose makes a larger percentage contribution to educational gradients at ages 30–60, ranging from 5.3 % to 6.4 % and 4.6 % to 8.1 % for males and females, respectively, in the earliest period. Increases in these contributions occurred in the late 1990s and were particularly large in the 2000s. In the most recent period, the contribution of drug overdose to life expectancy gradients reached upwards of 10 % for all education groups in the total population. This contribution was even larger among non-Hispanic whites. In 2007–2011, the contribution of drug overdose to educational gradients in life expectancy reached 15 % and 20 % for non-Hispanic white males and females, respectively.

Initially, the contribution of drug overdose to educational gradients in mortality was generally larger for males than females. Over time, however, drug overdose accounts for an increasingly greater proportion of educational gradients among females. This is true for both years of life lived between ages 30 and 60 and life expectancy at age 25, and it is somewhat more pronounced among non-Hispanic whites. Taking the groups with less than a high school education and college graduates as an example, we see that among non-Hispanic whites in 1992–1996, drug overdose accounted for 5.8 % of the gradient for males and 5.6 % of the gradient for females at ages 30–60. In 2007–2011, these figures were 15.4 % and 19.1 %, respectively.

### Contribution of Drug Overdose to Changes in Educational Gradients in Life Expectancy

Drug overdose makes sizable contributions to changes in educational gradients in life expectancy, and these contributions are larger for non-Hispanic whites than the total population. Between the periods 1992–1996 and 2007–2011, the difference in life expectancy at age 25 between the least- and most-educated non-Hispanic whites increased by 3.45 years for males and 4.27 years for females. Drug overdose accounted for 14.2 % and 11.9 % of these increases, respectively. Differences in life expectancy at age 25 between those with some college and those with college or more increased by 1.14 years for males and 0.65 years for females, with drug overdose accounting for 16.7 % and 19.3 % of these increases, respectively.

For males, the percentage contributions of drug overdose to changes in educational gradients in life expectancy remained fairly stable across time. For example, focusing on the change in the gradient in life expectancy at age 25 between non-Hispanic whites with less than high school and those with college or more, we see that contributions of drug overdose to the change between each of the three earlier periods and 2007–2011 are 14.2 %, 12.5 %, and 14.5 %, respectively. For women, the contributions to the changes between the first two periods (1992–1996 and 1997–2001) and the most recent period are fairly similar. However, the contribution of drug overdose tends to be much larger between 2002–2006 and 2007–2011.

For the population as a whole, longer-run trends in educational gradients in life expectancy between ages 30 and 60 have actually been quite favorable for men. In many cases, gradients decreased between 1992–1996 and 2007–2011. Between the two most recent periods, however, the gradients for less than high school and high school increased, with drug overdose accounting for 21 % of the 0.18-year increase in the gradient for high school graduates. Non-Hispanic white males experienced consistent increases in educational gradients in life expectancy at the prime adult ages. For all three education groups, drug overdose accounted for upwards of 70 % of the widening in the gradient between 1992–1996 and 2007–2011 as well as for 21 % to 45 % of the widening between the two intermediate periods and the most recent period.

For the female population as a whole, changes in educational gradients in life expectancy at ages 30–60 were fairly minimal. Where increases did occur, the contribution of drug overdose was sizable, upwards of 25 %. Non-Hispanic white females saw large increases in educational gradients in life expectancy at ages 30–60, with the largest increase observed for those with less than high school between 1992 and 2011. Drug overdose accounted for 42.3 % of this 0.47-year increase. Among non-Hispanic white females, drug overdose accounted for 26.4 % to 63.7 % of increases in educational gradients in life expectancy at these ages.

Finally, I examine two counterfactuals that further illuminate these results. Which scenario would result in greater gains in years of life lived, if all groups returned to their drug overdose mortality rates from the earliest period (prior to the drug overdose epidemic) or if they had the drug overdose mortality rates of the college-educated in each period?

Figure 2 shows the years of life that would be gained by education group in each of these two scenarios for males (panel a) and females (panel b). The escalation of the drug overdose epidemic is reflected in the increasing bar heights over time.

For males, the answer is clear: in each period, it is more favorable to have the drug overdose death rates of the college-educated in that period (i.e., the heights of the gray bars are greater than the black bars). However, the difference between the scenarios erodes over time. The difference in years of life gained between the two scenarios is largest in 1997–2001, the initial takeoff period for the drug overdose epidemic. In the two most recent periods, the years of life gained in the first scenario (the heights of the black bars) increase, and the years of life gained begin to converge between the two scenarios. There are some cases, usually for the some college group, where the years of life gained in each scenario are nearly equal.

For females, the story is similar to that for males in the earlier periods (all education groups would gain more years of life if they had the drug overdose mortality rates of the college-educated) but differs in the most recent period. In the most recent period, the high school and some college groups would gain more years of life if they reverted to their baseline drug overdose death rates than if they had the contemporaneous drug overdose death rates of college-educated females.

## Discussion

Since the mid-1990s, educational gradients in life expectancy widened, and dramatic increases in drug overdose mortality occurred. In this article, I show that drug overdose accounts for much of the widening in educational gradients in life expectancy between 1992 and 2011. Over this period, we observe a divergence between the college-educated and the three lower education groups, which experienced large increases in drug overdose mortality. The contribution of drug overdose to educational gradients in life expectancy is largest at the prime adult ages (ages 30–60) and for non-Hispanic whites.

The contribution of drug overdose to educational gradients in life expectancy at ages 30–60 is substantial and has doubled (for males) and tripled (for females) over time. In the most recent period, drug overdose accounts for roughly 16 % to 18 % of the difference in life expectancy between non-Hispanic white high school and college graduates aged 30–60. Among non-Hispanic whites aged 30–60, drug overdose accounts for more than 70 % of the widening in educational gradients in life expectancy among males and for 26 % to 62 % of the widening among females between 1992 and 2011.

A priori, one might expect the more-educated to have higher levels of drug use in the early stages of an epidemic: elites are typically on the vanguard of drug use (Collins 2005; Pampel 2001), and in the context of an epidemic driven by prescription painkillers, it is likely that the more-educated were among the first to be prescribed newly approved prescription opioids and better able to afford them. However, we do not observe higher drug overdose mortality among the college-educated in the context of the current drug epidemic. Instead, drug overdose mortality is higher and has increased most for the three lower education groups, suggesting that either use is lower among the college-educated and/or that the consequences of use in terms of mortality are very different for those with more versus less education.

Higher levels of drug overdose mortality among the less-educated may be driven by several factors. First, relative to the more-educated, the less-educated are more vulnerable to and have greater incentives to engage in drug use. In the face of deindustrialization and other structural changes in labor markets that lowered the returns to education levels below a college degree, the less-educated experienced declines in employment prospects, SES, and overall well-being (Kalleberg 2011; Weeden et al. 2007). Consequently, they are more susceptible to being targeted by sponsors who organize groups of users to obtain drugs from pill mills and have greater incentives to participate in the informal economies surrounding the diversion of OxyContin and other opioids (Keyes et al. 2014). Second, less-educated individuals face increased risk of workplace injuries, disability, and chronic health conditions (Minkler et al. 2006; Oh and Shin 2003; Seeman et al. 2008), which lead to greater likelihood of being prescribed opioid painkillers. Third, higher drug overdose mortality among the less-educated may be related to the geographic shift in drug overdose from cities to rural areas, especially in Appalachia and the Rust Belt. Rural areas have lower levels of educational attainment, experienced recent declines in socioeconomic well-being, and have poorer EMS infrastructure (Faul et al. 2015). The motivation for and returns to self-medication may be increased in such areas. Finally, the less-educated may have fewer resources to combat drug addiction, including financial resources, access to scarce slots in drug treatment programs, and support from social networks.

A particularly interesting finding of this study is the gender convergence in drug overdose. Historically, males had higher drug overdose death rates. In the course of the present epidemic, drug overdose mortality has converged between males and females, driven by more rapid increases among females. If we ask whether it is more favorable (in terms of years of life gained) for each group to return to their drug overdose death rates observed prior to the takeoff of the current epidemic or to have the drug overdose death rates of the college-educated, the answer differs for males and females. For males, it is always better to have the drug overdose mortality of the college-educated. For females in the most recent period, however, it is better for the some college and high school groups to return to their baseline levels than to have the contemporaneous death rates of the college-educated. This suggests that while educational gradients in drug overdose mortality remained fairly sharp for males, females experienced large increases in drug overdose mortality across all education levels over the course of the epidemic.

These gender differences may be due to differences in health-seeking behaviors and in social and biological dimensions of substance use and addiction. Men are more likely to use illicit drugs and alcohol, and women are more likely to engage in nonmedical prescription drug use (Simoni-Wastila et al. 2004). A study of all pharmaceutical overdose deaths that occurred in 2006 in West Virginia, which had the greatest increase in drug overdose mortality in 1999–2004, found that evidence of “doctor shopping” (having prescriptions for a controlled substance from five or more doctors in the year prior to death) was much more common among female (30.9 %) than male (16.7 %) decedents (Hall et al. 2008). This may be related to factors including women’s greater connectedness to the health care system and higher levels of stigma and social disapproval of illicit substance use for women than for men (Robbins 1989). Although overall rates of substance use are higher among males, females appear to have more rapid escalation of drug use, become addicted more quickly, and have more difficulty quitting after becoming addicted; on the other hand, women are also more likely to seek treatment for substance use addictions and to do so earlier than men (Becker and Hu 2008). Other contributing factors include gender differences in competing risks and trends in educational attainment. At the young and prime adult ages where drug overdose mortality is highest, males have higher death rates from other causes of death. Furthermore, educational upgrading has been more dramatic for females than males: among females, the less-educated—particularly the group with less than a high school education—are more select than corresponding groups among males.

### Limitations

Over time, education levels have increased in the United States, altering the population distribution across education groups (Hendi 2015). One consequence of this compositional change is that the less-educated have become increasingly select, affecting the comparison of education groups over time. The consequences of educational upgrading—such as worsening employment prospects and stagnating economic conditions for those with a high school diploma or less—are also likely to be key factors in the rise in drug overdose, and disentangling their separate influences is a complicated endeavor. This study does not account for these compositional changes, and caution should be taken in interpreting how drug overdose contributes to changes in educational gradients in life expectancy. The results pertaining to years of life lived between ages 30 and 60 are less sensitive to this issue because they make comparisons among individuals at most 30 years apart in age.

By necessity, this study relies on education reported on death certificates to estimate the fraction of all deaths due to drug overdose because sample sizes for drug overdose deaths based on survey data are too small to allow for stable estimates. To the degree that some error is introduced, it is introduced in both the numerator and the denominator of these fractions; that is, it is not subject to dual data source bias. This concern may be mitigated by the fact that education reporting on death certificates is considered to be more accurate at ages below 65 (Rostron et al. 2010), where the bulk of drug overdose mortality occurs.

Finally, drug overdose might be underreported on death certificates (see Paulozzi et al. 2006 for a fuller discussion). Such omissions would likely result in underestimates of years of life lost due to drug overdose, so the estimates in this article may be lower bounds. On the other hand, risky behaviors such as drug use, smoking, and alcohol use tend to cluster. It is possible that some drug overdose deaths had multiple contributing factors—or had these individuals not died of drug overdose, they would eventually have died of smoking- or alcohol-related causes. These possibilities might lead to overestimates of drug overdose mortality.

## Conclusion

This study documents increases in years of life lost due to drug overdose for all education groups and for both males and females between 1992 and 2011. The contribution of drug overdose to educational gradients in life expectancy increased over time, is larger at the prime adult ages (ages 30–60), and is greater for non-Hispanic whites than the population as a whole. Drug overdose accounts for a substantial proportion of changes in educational gradients between 1992 and 2011, particularly among non-Hispanic whites at ages 30–60. Over time, drug overdose mortality increased more rapidly for females than males, leading to a gender convergence.

These findings have important implications. Overall, they suggest that effective measures to address the drug overdose epidemic should take into account the differential burden across education groups. For example, drugs are increasingly being used to treat drug addiction. A recently proposed option of implants that release buprenorphine to treat opioid addiction must be replaced every six months by a trained doctor (Bebinger 2016). These and other addiction treatments being used are very expensive, and cost and access barriers for the less-educated are likely to be substantial. Finally, although the drug overdose epidemic was initially driven by prescription painkillers, it has moved on to encompass heroin and synthetic opioids such as fentanyl. To effectively address the epidemic and its contribution to educational gradients in life expectancy, we need to concentrate on the underlying factors that may be driving drug use, including poor social and labor market conditions. An important avenue for future research is to explore the linkages between gender and drug overdose, encompassing males’ initially higher mortality from drug overdose and the subsequent gender convergence.

## Acknowledgments

The author thanks Arun Hendi for helpful comments on earlier drafts of this article. The author is grateful to the National Center for Health Statistics for the use of the restricted-use microdata files for mortality. This research was supported in part by the Eunice Kennedy Shriver National Institute of Child Health & Human Development (NICHD) of the National Institutes of Health under Award Number K99 HD083519.

## Notes

1

This is related to the release of an abuse-deterrent reformulation of OxyContin in 2010 and significant increases in the availability of cheap, high-grade heroin throughout the United States.

2

In this article, I focus on the more-educated as one particular group of social elites.

3

These are deaths for which the underlying cause of death was ICD-9 codes E850–E858, E950.0–E950.5, E962.0, or E980.0–E980.5 prior to 1999 and ICD-10 codes X40–X44, X60–X64, X85, and Y10–Y14 from 1999 onward.

4

Fewer than 700 accidental poisoning deaths were recorded in the 1986–2004 NHIS files.

5

The latter measure is informative because it captures the ages at which drug overdose mortality rates are highest and have increased the most over time (see Fig. 3 in the  appendix, with the online version of the article showing the figure in color), and it is less sensitive to issues of age misreporting and changes in institutionalization at older ages.

6

This is partly related to differences in opioid prescribing by race/ethnicity. Blacks, Hispanics, and Asians are less likely than whites to receive opioid prescriptions, even controlling for pain severity (Burgess et al. 2014; Pletcher et al. 2008). Even if they are prescribed opioids, nonwhites are more likely to live in areas where they cannot obtain them. One study in New York City found that only 25 % of pharmacies in predominantly nonwhite neighborhoods carried sufficient supplies of opioids to treat severe pain compared with 72 % of pharmacies in predominantly white neighborhoods (Morrison et al. 2000).

## References

American Pain Society
. (
1999
).
Principles of analgesic use in the treatment of acute pain and cancer pain
(4th ed.).
Glenview, IL
:
American Pain Society
.
Bebinger, M. (
2016
).
FDA considering pricey implant as treatment for opioid addiction
.
Becker, J. B., & Hu, M. (
2008
).
Sex differences in drug abuse
.
Frontiers in Neuroendocrinology
,
29
,
36
47
. 10.1016/j.yfrne.2007.07.003.
Beltrán-Sánchez, H., Preston, S. H., & Canudas-Romo, V. (
2008
).
An integrated approach to cause-of-death analysis: Cause-deleted life tables and decompositions of life expectancy
.
Demographic Research
,
19
(
article 35
),
1323
1350
. doi:10.4054/DemRes.2008.19.35
Bohnert, A. S., Valenstein, M., Bair, M. J., Ganoczy, D., McCarthy, J. F., Ilgen, M. A., & Blow, F. C. (
2011
).
Association between opioid prescribing patterns and opioid overdose-related deaths
.
JAMA
,
305
,
1315
1321
. 10.1001/jama.2011.370.
Burgess, D. J., Nelson, D. B., Gravely, A. A., Bair, M. J., Kerns, R. D., Higgins, D. M., . . . Partin, M. R. (
2014
).
Racial differences in prescription of opioid analgesics for chronic noncancer pain in a national sample of veterans
.
Journal of Pain
,
15
,
447
455
.
Centers for Disease Control and Prevention (CDC), & National Center for Health Statistics (NCHS)
. (
2000/2003
).
Compressed mortality file 1979–1998
[CDC WONDER On-line Database, Series 20, No. 2A, 2000; & Series 20, No. 2E, 2003]. Retrieved from http://wonder.cdc.gov/cmf-icd9.html
Centers for Disease Control and Prevention (CDC) & National Center for Health Statistics (NCHS)
. (
2015
).
Compressed mortality file 1999–2014
[CDC WONDER Online Database, Series 20 No. 2T, 2015]. Retrieved from http://wonder.cdc.gov/cmf-icd10.html
Chang, H., Daubresse, M., Kruszewski, S. P., & Alexander, G. C. (
2014
).
Prevalence and treatment of pain in EDs in the United States, 2000 to 2010
.
American Journal of Emergency Medicine
,
32
,
421
431
. 10.1016/j.ajem.2014.01.015.
Chiang, C. L. (
1968
).
An introduction to stochastic processes in biostatistics
.
New York, NY
:
Wiley
.
Cicero, T. J., Ellis, M. S., Surratt, H. L., & Kurtz, S. P. (
2014
).
The changing face of heroin use in the United States: A retrospective analysis of the past 50 years
.
JAMA Psychiatry
,
71
,
821
826
. 10.1001/jamapsychiatry.2014.366.
Collins, R. (
2005
).
Interaction ritual chains
.
Princeton, NJ
:
Princeton University Press
.
Daubresse, M., Chang, H., Yu, Y., Viswanathan, S., Shah, N. D., Stafford, R. S., . . . Alexander, G. C. (
2013
).
Ambulatory diagnosis and treatment of nonmalignant pain in the United States, 2000–2010
.
Medical Care
,
51
,
870
878
.
Elo, I., & Preston, S. (
1996
).
Educational differentials in mortality: United States, 1979–1985
.
Social Science & Medicine
,
42
,
47
57
. 10.1016/0277-9536(95)00062-3.
Faul, M., Dailey, M. W., Sugerman, D. E., Sasser, S. M., Levy, B., & Paulozzi, L. J. (
2015
).
Disparity in naloxone administration by emergency medical service providers and the burden of drug overdose in US rural communities
.
American Journal of Public Health
,
105
(
S3
),
e26
e32
. 10.2105/AJPH.2014.302520.
Hall, A. J., Logan, J. E., Toblin, R. L., Kaplan, J. A., Kraner, J. C., Bixler, D., . . . Paulozzi, L. J. (
2008
).
Patterns of abuse among unintentional pharmaceutical overdose fatalities
.
JAMA
,
300
,
2613
2620
.
Hendi, A. S. (
2015
).
Trends in U.S. life expectancy gradients: The role of changing educational composition
.
International Journal of Epidemiology
,
44
,
946
955
. 10.1093/ije/dyv062.
Hendi, A. S. (Forthcoming).
Trends in education-specific life expectancy, data quality, and shifting education distributions: A note on recent research
.
Demography
.
Ho, J. Y. (
2013
).
Mortality under age 50 accounts for much of the fact that US life expectancy lags that of other high-income countries
.
Health Affairs
,
32
,
459
467
. 10.1377/hlthaff.2012.0574.
Ho, J. Y., & Fenelon, A. (
2015
).
The contribution of smoking to educational gradients in U.S. life expectancy
.
Journal of Health and Social Behavior
,
56
,
307
322
. 10.1177/0022146515592731.
Hoyert, D. L., Kochanek, K. D., & Murphy, S. L. (
1999
).
Deaths: Final data for 1997
(National Vital Statistics Reports, No. 47).
Hyattsville, MD
:
National Center for Health Statistics
.
International Narcotics Control Board
. (
2009
).
Report of the International Narcotics Control Board for 2008
(Report).
New York, NY
:
United Nations
.
Jonas, A. B., Young, A. M., Oser, C. B., Leukefeld, C. G., & Havens, J. R. (
2012
).
OxyContin as currency: OxyContin use and increased social capital among rural Appalachian drug users
.
Social Science & Medicine
,
74
,
1602
1609
. 10.1016/j.socscimed.2011.12.053.
Jones, C. M., Paulozzi, L. J., & Mack, K. A. (
2014
).
Sources of prescription opioid pain relievers by frequency of past-year nonmedical use: United States, 2008–2011
.
JAMA Internal Medicine
,
174
,
802
803
. 10.1001/jamainternmed.2013.12809.
Kalleberg, A. L. (
2011
).
Good jobs, bad jobs: The rise of polarized and precarious employment systems in the United States, 1970s–2000s
.
New York, NY
:
Russell Sage Foundation
.
Keyes, K. M., Cerda, M., Brady, J. E., Havens, J. R., & Galea, S. (
2014
).
Understanding the rural-urban differences in nonmedical prescription opioid use and abuse in the United States
.
American Journal of Public Health
,
104
(
2
),
e52
e59
. 10.2105/AJPH.2013.301709.
Klaue, K. (
1999
).
Drugs, addiction, deviance and disease as social constructs
.
Bulletin on Narcotics, LI
(
1–2
),
45
55
.
Kochanek, K. D., Murphy, S. L., Xu, J, Q., & Tejada-Vera, B. (
2016
).
Deaths: Final data for 2014
(National Vital Statistics Reports, Vol. 65 No. 4).
Hyattsville, MD
:
National Center for Health Statistics
.
Link, B. G., & Phelan, J. C. (
1995
).
Social conditions as fundamental causes of disease
.
Journal of Health and Social Behavior
,
35
(
Extra issue
),
80
94
.
McDonald, D. C., & Carlson, K. E. (
2013
).
Estimating the prevalence of opioid diversion by “doctor shoppers” in the United States
.
PloS One
,
8
,
e69241
. doi:10.1371/journal.pone.0069241.
Meara, E. R., Richards, S., & Cutler, D. M. (
2008
).
The gap gets bigger: Changes in mortality and life expectancy, by education, 1981–2000
.
Health Affairs
,
27
,
350
360
. 10.1377/hlthaff.27.2.350.
Miech, R., Pampel, F., Kim, J., & Rogers, R. G. (
2011
).
The enduring association between education and mortality: The role of widening and narrowing disparities
.
American Sociological Review
,
78
,
913
934
. 10.1177/0003122411411276.
Minkler, M., Fuller-Thomson, E., & Guralnik, J. M. (
2006
).
Gradient of disability across the socioeconomic spectrum in the United States
.
New England Journal of Medicine
,
355
,
695
703
. 10.1056/NEJMsa044316.
Minnesota Population Center, & State Health Access Data Assistance Center
. (
2015
).
Integrated Health Interview Series (IHIS): Version 6.12
.
Minneapolis
:
University of Minnesota
.
Morrison, R. S., Wallenstein, S., Natale, D. K., Senzel, R. S., & Huang, L. (
2000
).
“We don’t carry that”—Failure of pharmacies in predominantly nonwhite neighborhoods to stock opioid analgesics
.
New England Journal of Medicine
,
342
,
1023
1026
. 10.1056/NEJM200004063421406.
Muhuri, P. K., Gfroerer, J. C., & Davies, C. (
2013
).
Associations of nonmedical pain reliever use and initiation of heroin use in the United States
(CBHSQ Data Review report).
Rockville, MD
:
Substance Abuse and Mental Health Services Administration (SAMHSA)
.
Musto, D. F. (
1991
).
Opium, cocaine and marijuana in American history
.
Scientific American
,
265
(
1
),
40
47
. 10.1038/scientificamerican0791-40.
Oh, J., & Shin, E. H. (
2003
).
Inequalities in nonfatal work injury: The significance of race, human capital, and occupations
.
Social Science & Medicine
,
57
,
2173
2182
. 10.1016/S0277-9536(03)00073-X.
Okie, S. (
2010
).
A flood of opioids, a rising tide of deaths
.
New England Journal of Medicine
,
363
,
1981
1985
. 10.1056/NEJMp1011512.
Pampel, F. C. (
2001
).
Cigarette diffusion and sex differences in smoking
.
Journal of Health and Social Behavior
,
42
,
388
404
. 10.2307/3090186.
Pampel, F. C., Krueger, P. M., & Denney, J. T. (
2010
).
Socioeconomic disparities in health behaviors
.
Annual Review of Sociology
,
36
,
349
370
. 10.1146/annurev.soc.012809.102529.
Paulozzi, L. J., Budnitz, D. S., & Xi, Y. (
2006
).
Increasing deaths from opioid analgesics in the United States
.
Pharmacoepidemiology and Drug Safety
,
15
,
618
627
. 10.1002/pds.1276.
Paulozzi, L. J., Jones, C. M., Mack, K. A., & Rudd, R. A. (
2011
).
Overdoses of prescription opioid pain relievers—United States, 1999–2008
.
Morbidity and Mortality Weekly Report
,
60
,
1487
1492
.
Paulozzi, L. J., & Xi, Y. (
2008
).
Recent changes in drug poisoning mortality in the United States by urban–rural status and by drug type
.
Pharmacoepidemiology and Drug Safety
,
17
,
997
1005
.
Pletcher, M. J., Kertesz, S. G., Kohn, M. A., & Gonzales, R. (
2008
).
Trends in opioid prescribing by race/ethnicity for patients seeking care in US emergency departments
.
JAMA
,
299
,
70
78
.
Preston, S. H., Heuveline, P., & Guillot, M. (
2001
).
Demography: Measuring and modeling population processes
.
Oxford, UK
:
Blackwell Publishers
.
Rigg, K. K., March, S. J., & Inciardi, J. A. (
2010
).
Prescription drug abuse & diversion: Role of the pain clinic
.
Journal of Drug Issues
,
40
,
681
702
. 10.1177/002204261004000307.
Robbins, C. (
1989
).
Sex differences in psychosocial consequences of alcohol and drug abuse
.
Journal of Health and Social Behavior
,
30
,
117
130
. 10.2307/2136917.
Rostron, B. L., Boies, J. L., & Arias, E. (
2010
).
Education reporting and classification on death certificates in the United States
(Vital and Health Statistics, Series 2, No. 151).
Hyattsville, MD
:
National Center for Health Statistics
.
Rudd, R. A., Aleshire, N., Zibbell, J. E., & Gladden, R. M. (
2016
).
Increases in drug and opioid overdose deaths—United States, 2000–2014
.
Morbidity and Mortality Weekly Report
,
64
,
1378
1382
. 10.15585/mmwr.mm6450a3.
Seeman, T., Merkin, S. S., Crimmins, E., Koretz, B., Charette, S., & Karlamangla, A. (
2008
).
Education, income and ethnic differences in cumulative biological risk profiles in a national sample of US adults: NHANES III (1988–1994)
.
Social Science & Medicine
,
66
,
72
87
. 10.1016/j.socscimed.2007.08.027.
Simoni-Wastila, L., Ritter, G., & Strickler, G. (
2004
).
Gender and other factors associated with the nonmedical use of abusable prescription drugs
.
Substance Use & Misuse
,
39
,
1
23
. 10.1081/JA-120027764.
Sorlie, P. D., & Johnson, N. J. (
1996
).
Validity of education information on the death certificate
.
Epidemiology
,
7
,
437
439
. 10.1097/00001648-199607000-00017.
Temple, J. (
2015
).
American pain
.
Guilford, CT
:
Lyons Press
.
Warner, M., Chen, L. H., Makuc, D. M., Anderson, R. N., & and Miniño, A. M. (
2011
).
Drug poisoning deaths in the United States, 1980–2008
(NCHS Data Brief, No. 81).
Hyattsville, MD
:
National Center for Health Statistics
.
Weeden, K. A., Kim, Y., Di Carlo, M., & Grusky, D. B. (
2007
).
Social class and earnings inequality
.
American Behavioral Scientist
,
50
,
702
736
. 10.1177/0002764206295015.
Xu, J. Q., Murphy, S. L., Kochanek, K. D., & Bastian, B. A. (
2016
).
Deaths: Final data for 2013
(National Vital Statistics Reports, Vol. 64, No. 2).
Hyattsville, MD
:
National Center for Health Statistics
.